[pulseaudio-discuss] [PATCH 01/13] cpu: Add force_generic_code flag to cpu_info struct

[David Henningsson]

On 2014-09-10 16:28, Peter Meerwald wrote:
> Hello David,
>
>>> The remapper and channel mixing code have (faster) specialized and (slower)
>>> generic code certain code path. The flag force_generic_code can be set to
>>> force the generic code path which is useful for testing. Code duplication
>>> (such as in mix-special-test) can be avoided, cleanup patches follow.
>>>
>>> Signed-off-by: Peter Meerwald <pmeerw at pmeerw.net>
>>> ---
>>>    src/Makefile.am       |    2 +-
>>>    src/daemon/main.c     |    4 +++-
>>>    src/pulsecore/cpu.c   |   28 ++++++++++++++++++++++++++++
>>>    src/pulsecore/cpu.h   |    5 +++++
>>>    src/pulsecore/mix.c   |    8 ++++++++
>>>    src/pulsecore/remap.c |   13 +++++++++++++
>>>    6 files changed, 58 insertions(+), 2 deletions(-)
>>>    create mode 100644 src/pulsecore/cpu.c
>>>
>>> diff --git a/src/Makefile.am b/src/Makefile.am
>>> index 51ef690..634e26b 100644
>>> --- a/src/Makefile.am
>>> +++ b/src/Makefile.am
>>> @@ -897,7 +897,7 @@ libpulsecore_ at PA_MAJORMINOR@_la_SOURCES = \
>>>    		pulsecore/rtpoll.c pulsecore/rtpoll.h \
>>>    		pulsecore/stream-util.c pulsecore/stream-util.h \
>>>    		pulsecore/mix.c pulsecore/mix.h \
>>> -		pulsecore/cpu.h \
>>> +		pulsecore/cpu.c pulsecore/cpu.h \
>>>    		pulsecore/cpu-arm.c pulsecore/cpu-arm.h \
>>>    		pulsecore/cpu-x86.c pulsecore/cpu-x86.h \
>>>    		pulsecore/cpu-orc.c pulsecore/cpu-orc.h \
>>> diff --git a/src/daemon/main.c b/src/daemon/main.c
>>> index 02a8ea6..74527be 100644
>>> --- a/src/daemon/main.c
>>> +++ b/src/daemon/main.c
>>> @@ -1050,13 +1050,15 @@ int main(int argc, char *argv[]) {
>>>    #endif
>>>
>>>        c->cpu_info.cpu_type = PA_CPU_UNDEFINED;
>>> +    c->cpu_info.force_generic_code = false; /* use generic, slow code */
>>
>> Are you sure this comment is correct? It looks like the opposite.
>
> the comment is not very clear at least; it describes the purpose of the
> flag, not the assignment
>
> replacing it with
> /* don't force generic code, used for testing only */
> in v2
>
>> Also, what is the difference between setting force_generic_code and setting
>> PULSE_NO_SIMD?
>
> mixing and remapping as generic and special-case code; the flag can be
> set to force use of the generic code
> this is useful for the test code so that special-case code can be compared
> with the (presumably correct) generic code -- it saves quite a lot of code
> in the test suits (the alternative would be to expose the internal
> mixing/remapping function); special-case code path are beneficial on all
> CPUs
>
> PULSE_NO_SIMD suppresses CPU-specific code path

I first confused "special-case code" with "CPU-specific code", but is 
this correct:
"special-case code" is when you e g have a special function for "mono to 
stereo" conversion instead of the generic "m to n channels" conversion,
whereas "CPU-specific code" means code that uses special instructions 
(usually hand written assembly)
And then there is orc, which seems to count as "CPU-specific code" in 
this context.

I guess the "force_generic_code" name was too generic for me to 
understand :-) maybe two bitflags would be even better, e g 
"use_cpu_specific_code" and "use_special_case_code", or something...

-- 
David Henningsson, Canonical Ltd.
https://launchpad.net/~diwic